Between model-based and visuo-motor motion control

Habilitation committee¶

• François Chaumette, Senior Scientist at Inria (president)
• Serena Ivaldi, Senior Scientist at Inria (examiner)
• Abderrahmane Kheddar, Senior Scientist at CNRS (examiner)
• Leonardo Lanari, Professor at the University of Rome (examiner)
• Marc Lelarge, Senior Scientist at Inria (examiner)
• Katja Mombaur, Professor at KIT (examiner)
• Quang-Cuong Pham, Associate Professor at NTU (examiner)
• Ludovic Righetti, Associate Professor at NYU (reviewer)
• Philippe Soueres, Senior Scientist at CNRS (reviewer)
• Eiichi Yoshida, Professor at the Tokyo University of Science (reviewer)

Abstract¶

In this presentation, we first discuss the notion (desired, but undefined) of "stability" for robot locomotion. After some physics reminders on the zero-moment point (ZMP) and its corresponding support area, we see how these concepts generalize to systems that make multiple contacts with their environment. Teaser: the support areas become support volumes over more general quantities, prompting our interest into polyhedral projection. We review algorithms for it, including output-sensitive contour tracking and its lazy sibling output-sensitive ray casting.

We then turn our gaze to bipedal locomotion with a focus on big robots climbing stairs. We explore the line of ideas around divergent components of motion (DCM) for some first viability guarantees. We then dive into a "space-time" 4D DCM that generalizes the standard one to account for rough terrains, and see how complex behaviors like choosing a balancing strategy can emerge from simple linear feedback of this 4D DCM.

The algorithms at our disposal for optimizing locomotion trajectories having evolved significantly, the last part of the presentation shifts to visuomotor control, that is, control based on partial and ambiguous observations of an environment whose state is no longer assumed to be known. It mainly describe the axes of my current project in this direction.

Content¶

Slides

Bio¶

Stéphane is a research scientist at Inria. He received his M.Sc. in computer science from the École Normale Supérieure in 2012, and his Ph.D. in mechano-informatics from the University of Tokyo in 2016. Stéphane has worked at CNRS as tenured researcher and at ANYbotics AG as locomotion team lead before joining Inria where he is currently (having a blast) bridging motion control and computer vision. Stéphane is a proponent of open source robotics and contributes to projects like Upkie wheeled bipeds, robot_descriptions.py and qpbenchmark.